Download Fraud prevention

Journal of Advanced Analytics
3Q 2O12
2 Fraud is everyone’s problem
4
Fraud detection is more than cool
7
LA County closes in on fraud rings
9
Preventing fraud for the good
of society
11
Is your health care fraud detection
solution ignoring valuable sources
of data?
13
Hyundai Marine & Fire Insurance
prevents fraudulent auto claims
15
The future of fraud investigations
17How to prevent fraud in the Indian
telecom industry
20 Cutting tax errors in half
22Reducing tax fraud leads to better
customer service in Belgium
25Using Analytics to Predict Fraud
fraud
prevention
Learn how leading organizations and
institutions prevent fraud in the digital age.
Journal of Advanced Analytics
what’s in this issue
3Q 2O12
Editorial Director
Mikael Hagstrom
[email protected]
Banks, insurance companies, health
care organizations and government
entities are all seeing an increase in the
incidence and sophistication of fraud,
waste and abuse activities, fueled in
large measure by the financial turmoil
gripping the world’s economy. To fight
fraud effectively, organizations must
continually improve the monitoring
of customer behavior across multiple
accounts and systems. Combining
technologies like business rules,
anomaly detection, predictive and
learning models, and social network
analysis allows organizations to
predict fraud with astounding
accuracy and take steps to prevent
fraud from occurring.
Editor-in-Chief
Alison Bolen
[email protected]
Managing Editor
Anne-Lindsay Beall
[email protected]
Copy Editors
Amy Dyson
Chris Hoerter
Amy Madison
Trey Whittenton
Intelligence Quarterly is published quarterly by SAS Institute Inc. Copyright
© 2012 SAS Institute Inc., Cary, NC, USA. All rights reserved. Limited copies
may be made for internal staff use only. Credit must be given to the publisher.
Otherwise, no part of this publication may be reproduced without prior written
permission of the publisher.
SAS and all other SAS Institute Inc. product or service names are registered
trademarks or trademarks of SAS Institute Inc. in the USA and other countries.
® indicates USA registration. Other brand and product names are trademarks
of their respective companies.
SAS is the leader in business analytics software and services, and the largest
independent vendor in the business intelligence market. Through innovative
solutions, SAS helps customers at more than 60,000 sites improve performance
and deliver value by making better decisions faster. Since 1976 SAS has been
giving customers around the world THE POWER TO KNOW.
®
Editorial Contributors
Michael Dowding
John Geurts
Amitava Ghosh
Eva Gustin
David Hartley
Graham Kemp
Eva (Hyun-min) Kim
Julie Malida
Marcie Montague
Mark Moorman
Ramesh Subrahmanyam
David Stewart
Amar Vohra
Katrina Wakefield
Art Direction
Brian Lloyd
Photography
John Fernez
Steve Muir
2
Fraud is
everyone’s problem
You can put a stop to it with predictive capabilities
and a holistic framework for fraud
Mikael Hagstrom Executive Vice President, SAS
In March 2012, Nikolay Garifulin was
sentenced to two years in prison for his
part in the Zeus cybercrime spree that
defrauded banks for hundreds of millions
of dollars and infected more than 2,400
computers around the world. Garifulin
himself was just one part of a franchise
of fraudsters that used the Zeus malware,
which allowed an imposter to capture
personal information from the computers
it infected. Garifulin was convicted of
stealing more than $3 million.
As a point of comparison, Thomas
Woodward was sentenced to 10 years in
prison for robbing $4,267 from a bank in
Massachusetts. While Garifulin’s sentence
was considerably lighter, the most interesting thing, frankly, is that he was sentenced
at all. Most fraud and abuse activity goes
untried. In many cases, it goes undetected.
Typically, the cost of fraud is passed on
to customers, constituents or owners in
the form of higher fees, increased taxes or
lower margins.
What the Zeus cyberattack and others
like it teach us is that fraudsters are
prevalent, active, coordinated and
technically skilled. In our efforts to open
our organizations to Web commerce,
we have created new concerns for the
protection of our funds, information
and customer data. Cyberattacks will
continue, and some believe there may
even be a crippling attack in the near
future, which could lead to a crisis with
even more devastating impact than our
current financial crisis.
It doesn’t have to be this way. In fact, the
digital age has more to offer the defenders
than the offenders. With analytics, you can
move frontline fraud management from
detect-and-defend to prevent-and-eradicate.
While many still engage in an arms race
of rule-based detection to lock offenders
out, this issue of the Intelligence Quarterly
highlights those who analyze big data to
neutralize organized crime and to prevent
the opportunistic seeker from making the
wrong choice.
Own and understand
the fraud process
Fraud is everyone’s problem and cannot
be ignored. Its cost is trickled down through
the economy, putting pressure on those of
us who are honest and forthright in our
dealings. For many people in government
and business today, it is difficult to think
like a fraudster, but just one abusive
person in 100 can make all the difference.
These social vampires exist, and they
are collaborating.
3
Too often, I see organizations with limited
fraud prevention, such as a black box
system – or worse, none at all, other than
a firewall. As more and more doors open
through growing online services, it is
imperative that your organization’s fraud
prevention efforts keep up.
Fortunately, high-performance analytics,
coupled with big data, is available to
organizations susceptible to fraud and
abuse, helping them manage and prevent
these activities. Many organizations around
the world are using data and analytics to
build fraud prevention technologies that
save money, protect corporate reputation
and reduce process costs. It is important
not to lock yourself into a black box fraud
solution built on some historical perspective
of fraud. Instead, success comes when
you can own and understand the fraud
process and respond to the ever-changing
dynamics of fraudsters.
A holistic approach
to fraud management
Fraud has often been compared to a
balloon, since pressing on one place in the
balloon just forces the air into another. Like
the air, fraud moves from one inefficient
process to another within an organization.
When we stop transactions or decline
claims without prosecuting the person
responsible and without fixing the
inefficient processes to begin with, we
are training fraudsters to just keep trying.
This cycle teaches people who are bent
on criminal behavior how to attack your
system. Many times you need to follow the
suspect, rather than just disconnecting, in
order to convict and eradicate. Indeed,
business changes can often have a big
impact on your fraud exposure, making
it essential to be active and elastic in
fraud prevention – not just fraud detection
– and to use technologies that will grow
as your needs grow.
A fraud framework is a complete set of
processes that access and integrate data,
produce alerts, provide holistic reporting,
control workflows and case management,
and learn from past experience to become
– and remain – effective.
Organizations from the public and
private sectors alike are finding this fraud
framework valuable. We introduce you
to some of these organizations in this
issue, including:
• The Swedish tax office, where fraud
prediction has changed the entire tax
collection process – and improved the
public’s perception of paying taxes
(Page 20).
• LA County, where analysis of social
networks, combined with other fraud
predictive methods, has prevented
millions of dollars from going to childcare
fraud rings, money that would otherwise
have been stolen from the defenseless
with the greatest need (Page 7).
• Hyundai Marine & Fire, whose fraud
prediction and prevention programs help
protect the company’s most valuable
customers (Page 13).
Also in this issue is an explanation of
how other tax revenue offices use a fraud
framework to monitor value-added-taxes
(VAT) for fraudulent activities, such as
carouseling, by engaging in preventive
services that help reduce errors in tax
returns. Banks, health care providers and
welfare organizations have all used this
framework to protect themselves from
fraudulent and abusive behavior.
Along with accurate detection, these
organizations find value in a holistic
approach to fraud management.
Combining technologies such as
business rules, anomaly detection,
predictive and learning models, and
social network analysis can predict
fraud with astounding accuracy. Using
analytic models like neural networks,
these systems become learning systems
that get better over time, and they become
more dynamic by accounting for evolving
fraud activities and by using anomaly and
social network models.
Fraud continues to be a major concern and
that concern continues to grow. Currently,
there are more white collar workers
unemployed than ever before – and fraud
is a white collar crime. The “fraud triangle”
describes three factors present in fraud:
motive, rationalization and opportunity.
With a highly skilled unemployed
population, these causal factors make
for a trained, motivated and potentially
desperate group of people. Add to that
the Internet access available to so many
people today and you have a toxic mix.
Most governments and organizations
around the world are beginning to take
this growing threat seriously. Now is
your chance for your organization to
aggressively pursue fraud prevention
processes. I hope this issue of Intelligence
Quarterly will motivate you to take that
first step toward a better future.
online
Read Mikael Hagstrom’s blog:
blogs.sas.com/mikaelhagstrom
As head of an expanding global team of 4,500 professionals in 48 countries, Mikael
Hagstrom is passionate about providing a culture where innovation can flourish,
resulting in market leadership for the organization and its customers. He leads SAS’
Europe, Middle East, Africa (EMEA) and Asia Pacific regions, which account for 54
percent of SAS’ 2011 revenue, or $1.47 billion.
4
Fraud detection is
more than cool
On financial crime, capable guardianship
and behavioral analytics
John Geurts Chief Security Officer, the Commonwealth Bank Group
You can’t deny the cool factor of fraud
prediction. Today, at the Commonwealth
Bank Group, for those products and
channels we monitor, we can reliably
predict the likelihood of fraud activity for
any given transaction before it is authorized. Not only that, but we can do it up to
250 times per second, get answers within
40 milliseconds of the transaction being
initiated, without transaction sampling,
and across multiple channels and products.
challenge in a complex technological
age where the range of products and
services has evolved to meet the needs
and demands of our customers.
However, it is worth understanding the
principles that formed the basis of how we
got here and also ask, “Can we do more?”
Are you and I partly
accountable for fraud?
To truly understand the challenge, we
can draw inspiration from research that
pre-dates consumer e-commerce. Criminologists Lawrence E. Cohen and Marcus
Felson introduced the concept of capable
guardianship, which identifies the role we
all play to address the three precursors to
crime: motive, means and opportunity.
Fraud in the banking industry is not just
another cost of doing business that needs
to be priced. It has its roots in crime,
and we must defend ourselves against
crime and misadventure. This can be a
They contend that crime is normal and that
it depends on the opportunities available.
If the target is not protected, crime will
happen. If this is the case, then our job in
financial services is more difficult because
we are not just protecting ourselves from
criminals or hardened offenders. We must
also protect ourselves from people who
take advantage of an opportunity that
we created by failing to provide
capable guardianship.
As a provider of services, we must take
accountability for fraud and use the
resources we have available to anticipate,
prevent, detect and respond to fraud if it
occurs. Capable guardianship implies that
we, as a key component in the chain of
criminal behavior in banking, must invest
both intellectually and practically in guarding everything within our power to protect.
Predicting when
fraud will occur
Of course, the opportunities to commit
fraud have changed with the advent of
5
technology. As systems and processes
become more automated, we have an
increased opportunity to implement systems
and controls at appropriate places in the
cycle to prevent and detect fraud.
What is a
multilayered
approach
to fraud?
In building their fraud detection methods
and models, firms are using all of the
following approaches:
• Business rules. Individual rules score
or define alerts based on intuition and
general experience.
• Anomaly detection. Alerts are
defined based on events that
represent statistical deviations
from normal or expected behavior.
• Predictive models. Full-scale statistical
models establish alerts based on a
risk score derived from event
characteristics that are indicators
of prior fraud incidents.
• Social network analysis. Alerts are
based on the level of association
(through shared or similar attributes)
between the current event and
individuals or accounts that are known
or suspected of fraudulent behavior.
The choice of which methods to use often
depends on the details of the application
and institution. In general, there is a trend
away from using business rules as the sole
method for defining alerts, and a broader
trend toward using all methods as needed.
Our key principle in detecting fraud is,
simply, to stop looking for fraudulent
transactions. That sounds counterintuitive,
but it actually works.
Instead, our approach is to predict whether
or not an event or transaction has occurred
that will give rise to fraudulent activity,
money laundering or other proscribed
activities. This approach does not mean the
end of transaction fraud rules, but they do
have their limitations.
Events to watch include customer and
network activities, account transactions
or activities surrounding entire classes
of accounts. In essence, we are looking
for behavioral indicators of fraud or
other activity.
It is important that we choose “the right
time” to interdict a transaction of interest.
We should not be seduced by the need to
do everything in “real time” unless we need
to, because real time is expensive.
For the Commonwealth Bank, fraud
detection needs to be real-time, as we
have a real-time core in our retail and
business banking platform, and we
seek to determine whether or not a
transaction is fraudulent before it is
authorized. On the other hand, an
overnight batch is more appropriate
for anti-money laundering obligations.
Likewise, when detecting payments to
sanctioned countries and entities, we can
delay that transaction to make a decision
later that day.
Fraudulent behavior is
inconsistent with
normal behavior
The real difference that has emerged
in our thinking in the past six years is
that we stopped just looking for fraud
at a transaction level and started
looking for changes in our customers’
overall behavior.
Criminals do not segment themselves
by product or service or geography.
What they are actually doing when
committing fraud or laundering money
is taking advantage of a weakness,
more often involving a customer or the
customer’s data. The fraudulent act is
a behavior that can be recognizable
through advanced modeling techniques
because we can anticipate that the
behavior is sufficiently inconsistent
with known normal behavior.
This change in our thinking, coupled with
a desire to streamline a number of siloed,
product-specific fraud detection platforms,
led to the use of SAS.
®
SAS had a longstanding relationship
with the Group through credit risk and
customer marketing analytics solutions.
In fact, much of our early success in
fraud detection was due to two skilled
analysts from our marketing department
who joined our team and developed
our early lending and transaction fraud
models. Their approach was to identify
the range of data sources that would be
required to build a behavioral profile of
the customer.
System requirements
for fraud detection
Beyond the increased possibilities presented by the behavioral analytics approach,
how does the approach work from a
practical perspective?
Based on a customer-centric model, consider this example: A customer withdraws
money with a chip-enabled debit card at
an ATM in her home city, lends a friend
$50 using her mobile banking device and
uses Internet banking to transfer funds to
a third party from an IP address located
offshore. The Internet banking transaction
6
is most likely to be fraudulent, and we must
be able to pass a message back through
our system to deny the transaction within
40 milliseconds or so.
Without a customer-centric fraud detection
system, we likely would not be able to
identify the transaction as fraud in time
to stop it, and we probably would not
even know it occurred until the customer
complained some weeks later.
Fundamentally, fraud systems for commonly
used retail products and channels must be
linked to customer behavior, and we
must have a good idea what “normal
banking activity” looks like; otherwise,
the opportunities to detect and prevent
fraud are lessened.
This is not a trivial task. The efforts required
to detect fraud, money laundering and
other proscribed activities demand a
disciplined approach and robust systems.
The Commonwealth Bank operates two
SAS platforms to attend to the majority of
real-time and batch requirements for both
fraud and money laundering.
The batch platform, which we call the
Financial Crimes Platform, was developed
in 2006 for transaction and origination
fraud before it was extended successfully to
money-laundering detection. The real-time
platform was launched in 2011 for the
Group’s extensive debit card portfolio,
with the migration of other channels and
products from legacy systems progressing.
In order to understand the scale of the
systems, consider the following facts.
transactions every night, looking for
fraud and money laundering activity.
The Real-Time Transaction Monitoring
system (to prevent fraudulent transactions
in real time) includes:
• 11 million account profiles.
• 6 million customers.
• A current average of 80-85 transactions
per second with a mean response time of
40 milliseconds.
• Tested peak volume of 250 transactions
per second.
Combining a customer-centric view of fraud
with advanced analytics and computing
capabilities presents many opportunities.
From a traditional fraud perspective, our
systems can help us:
• Identify fraud where the customer is the
victim or the perpetrator.
• Identify activity that gives rise to the
Group submitting a suspicious activity
report pursuant to our statutory antimoney laundering obligations.
customer-centric approach to analytics
is marketing.
Declining non-fraud transactions for
valued customers provides a suboptimal
service experience. However, we mitigate
that risk by understanding all transactions
and building a detailed view of what is
considered normal. The factors we use to
determine what is normal should also be
applicable when we consider how to
generate just-in-time marketing leads
without swamping our consumers with
too many leads.
We can do more – I have no doubt of
that. While our primary role is to ensure
the fraud detection systems are optimized
and applicable to the threats we face, we
should take every opportunity to leverage
our investment in advanced systems to
improve our return on investment.
online
Read more from John Geurts:
sas.com/success/CBA.html
• Provide an opportunity (in the case of a
false positive) to learn more about our
customer banking behavior.
Can we do more?
We have demonstrated the trust we place
in our systems to reliably defend us and our
customers from criminal activity.
What we are yet to demonstrate, but
intuitively believe is probable, is that the
fourth opportunity to be gained from a
The Financial Crimes Platform
(used to detect fraud and money
laundering) includes:
• 31 source systems.
• 11 million customers.
• 15 million transactions
loaded each day.
• Analyses of up to 420 million
John Geurts is the Executive General Manager for Group Security and Chief Security
Officer for the Commonwealth Bank Group in Australia, and has had the pleasure
of leading Group Security for almost 12 years. Group Security provides global
security leadership for the bank, including its international subsidiaries and
majority joint ventures.
7
LA County closes in
on fraud rings
Identifying child care fraud sooner
saves millions per year
Los Angeles (LA) County has the largest
population of any county in the United
States, with approximately 27 percent of
California’s residents. If it were a country,
LA County would be among the top
20 in the world as measured by GDP.
The county’s budget is approximately
$23 billion, which goes toward
providing social, health-related and
law enforcement services.
Fraud Framework and SAS Social
Network Analysis to get a better handle
on the problem.
Recently, LA County has been challenged
by an increase in fraud related to child care
services. The county estimates that fraud has
grown by about 40 percent and, in many
cases, is perpetrated by highly organized
fraud rings. To stem the growing tide of
financial crimes and losses, the Chief
Executive Office of LA County and the
Department of Public Social Services
conducted a proof of concept (POC) to
assess the feasibility of using the SAS
“Using the results of the POC we did
projections and cost-savings analysis and
concluded that the county could expect to
see a return on investment of somewhere
between $7 million to $30 million
annually,” says Manuel Moreno, Director
of Research, Chief Executive Office. “The
POC showed a number of advantages in
using the technology, not only related to
costs, but how we prioritize investigations.
®
®
The Chief Executive Office is responsible
for establishing and administering the
budget for the county’s services and
programs, which includes monitoring
and measuring them for efficiency and
cost-effectiveness.
8
“The county [expects] to see a return
on investment of somewhere between
$7 million to $30 million annually.”
Manuel Moreno, PhD, Director of Research
With social network analysis and
predictive analytics, we can identify
benefit recipients and service providers
and predict those most likely to engage
in fraud and where there might be a
potentially large loss of funds.”
Moreno says common fraud cases are
characterized by criminals filing false
claims of employment, which usually
include declaring employees who don’t
exist. In some cases, businesses are set
up by the heads of fraud rings. These
individuals are in collusion with recipients,
who falsely declare that their children are
attending nonexistent child care programs.
Sometimes, criminals declare work
schedules that are either false or shorter
than the amount of time claimed.
The POC demonstrated that by using
predictive models and peer group
analysis to detect behavioral anomalies
in the utilization of child care services,
fraud risk scores could be developed to
help decrease the number of false positive
cases assigned to investigators. The POC
also allowed the team to map potential
fraud rings.
“One of the things we showed is the
system’s ability to utilize social networks
to detect if individuals are likely to
commit fraud, based on their fraud
risk score,” explains Moreno. “We
established a network, for example,
consisting of participant and provider
nodes and illustrated their relationships.
We looked at whether this small network
fit into the larger scheme of networks, in
which participants are in collusion with
other child care providers. With the
technology, we identified strong central
nodes. In one case, we had a child care
provider servicing many nodes of participants engaging in collusion activities.
“We needed a unified approach to get a
handle on the fraud problem,” Moreno
continues. “We wanted a solution that
provided data integration, as well as a
powerful analytical tool workbench. To
perform sophisticated predictive modeling
we needed historical data, and for that
we had to integrate many external and
internal data sources, such as the state
of California’s Employment Development
Department data and business license
data from Los Angeles County.”
Based on metrics from the POC,
Moreno and his team concluded that
data integration, predictive capabilities,
data mining tools and the insight
by the SAS solution can be used
effectively to detect fraud before
it occurs.
“We calculated that the accuracy rate
of fraud rings identified by the social
network analysis solution to be, with
reliability, 85 percent.”
online
Learn more about
SAS Social Network Analysis:
sas.com/solutions/fraud/
social-network
What if
you could...
• Improve information credibility.
What if you could easily eliminate
duplicate names, addresses and other
identifying information from your data
to reduce erroneous payments and
duplicate billing?
• Enhance audit effectiveness. What if
you could predict the likelihood that a
transaction would be fraudulent and flag
suspicious activity for further investigation
so you could not only uncover fraud, but
prevent fraudulent payments?
• Deliver fact-based insight. What if you
could not only access reports that
measure your efforts toward reducing
fraudulent payments, but also drill down
for details at the department level?
• Achieve financial accountability.
What if your financial system could
track program purchases, payments
and costs to ensure that they are
necessary and justified?
• Manage performance. What if everyone
in your organization worked together
collectively and collaboratively, sharing
knowledge and best practices, to
achieve common goals for preventing
or correcting improper payments?
You can with the SAS Fraud Framework:
sas.com/solutions/fraud/index.html
9
Preventing fraud for
the good of society
We can no longer afford to let fraud and error burn a
hole in our pockets
Graham Kemp Public Sector Head, SAS UK
Tackling fraud is seen not only as a
moral issue nowadays, but also as the
most obvious, and possibly the least
painful, way to reduce deficits. In April,
the UK’s National Fraud Authority
released the results of its Annual Fraud
Indicator, which found that fraud against
the public sector has been revised down
from US$33.2 billion to US$31.8 billion
per annum, influenced to a large extent
by reduced fraud against the tax system.
While the figures are clearly a step in the
right direction and show that measures
the government is putting in place are
beginning to take shape, the government
cannot afford to rest on its laurels. If the
public sector is to build on the momentum of the announcement, it will need
to ensure that civil servants have the
proper training and procedures in place
to continue to prevent, detect and punish
fraudulent activity effectively.
Fraud has long been associated with
lack of effective monitoring of financial
information and transactions. In the
private sector, the insurance industry is
already implementing best practices that
collect, analyze and share information
between otherwise competing partners
(see Page 15).
Collect, analyze and share
To combat fraud, the public sector needs
to improve efficiency by using an intelligent approach to data analytics to detect
abnormal patterns, link multiple parties,
automatically route suspicious cases for
further investigation, and use predictive
modeling to uncover new fraudulent
activity. Government departments must
also accelerate cross-government working
and adopt a data-sharing policy where
appropriate.
Public sector organizations must recognize
that dealing with fraud in its various forms,
whilst balancing increased pressure for
savings, can only be achieved by treating
information as an asset that should be
collected and analyzed in smarter ways.
As is often the case, you do not know
whether you have the right information
until you act upon it. It is also true that
different public sector stakeholders have
different views on what information could
be collected, shared and analyzed. The
information accessible to public sector
organizations includes both structured
and unstructured data – such as text,
video, audio and social networking
1
0
information. To better manage this,
comprehensive frameworks that support
integration of the different types of
information need to be developed.
results were slightly better, but 44 percent
of respondents still lacked awareness of
their agency’s or department’s approach
to the issue.
Irish Revenue
Commission
uses SAS to
target effective
interventions
®
Develop best practice guidelines
As it stands, fraud and error represent
a huge black hole in the government’s
balance book. If the government wishes
to realize its vision of an all-pervasive,
sustained, zero-tolerance culture to fraud
and error across the public sector, civil
servants will first need to better understand
exactly how their departments are already
tackling fraud and error, and have the
right training and incentives to ensure that
they can build on this.
The government is already using software
solutions and predictive analytics as a
key enabler to combat fraud, but skills
and training are still a serious issue.
SAS recently polled civil servants on this
issue and found that 73 percent of those
questioned claimed to have received no
training in tackling fraud and error over
the last 12 months. It is imperative that the
government addresses this if it wishes to
root out fraud across the board.
The findings from our research suggest
that, despite the billions of pounds lost
to fraud and error every year, the level
of awareness about how public sector
departments were addressing the issue
is low. Nearly half of those questioned
were unsure whether their department
had carried out an investigation into fraud
over the last 12 months, and this was only
slightly lower among the senior-grade
sample. For investigations into error, the
The time to act is now
The UK public sector needs to be
innovative in order to protect the citizen
from the effects of fraud, both individually
and also to ensure that taxpayers’ money
is not used to pay fraudulent claims.
Otherwise, confidence in the public sector
will be eroded and money will be taken
from the pockets of the most needy. The
fraudsters aren’t going to give up anytime
soon, so the onus remains on the government to take the fight to them.
online
Learn how to combine traditional fraud
detection methods with modern methods
like social network analysis:
sas.com/fraud-hybrid
The Irish Revenue Commissioners (IRC)
worked with SAS to develop predictive
models that moved from identifying risks
of non-compliance or liquidation, to
predicting the likelihood of a case yielding in the event of an intervention, such
as an audit. The models also assess the
value of a potential yield to the IRC, so
that interventions can be prioritized and
tailored accordingly. Early results have
been positive, with a strong relationship
between cases predicted to yield and
actual average yield. See Page 25
to learn more about how IRC uses
analytics to prevent fraud.
Fraud tips webcast:
sas.com/fraud-webcast
Graham Kemp has more than 10 years of experience advising public sector organizations
on how to select and deploy effective technology solutions to meet their unique needs. He
participated in the Government’s Information Age Partnership. He is also a former chair
of the Department for Education ICT Industry Club, Infrastructure Group and a member of
the Government’s Distance and Education Learning Group advisory board.
1
1 Is your health care
fraud detection
solution ignoring
valuable sources
of data?
Julie M. Malida Principal for Health Care Fraud, Enterprise Financial Crimes Practice, SAS
In the health care industry, professional
fraud often involves multiple parties that
are in collusion, premeditated schemes,
identity theft and organized crime.
Frequently, the money evaporates in a
matter of days. Consequently, it is now
more important than ever to be timely
in fraud detection (preferably before
monies are paid out), employ multiple
analytical methods of detection, and use
a variety of data sources beyond just
claims data.
Let’s take a closer look at the most common
methods of fraud detection and the
frequently ignored data sources that
can help improve detection.
Multiple methods of analysis
Fraud is a spectrum of activity that
ranges from opportunistic deception to
truly premeditated, organized schemes.
Each different type of analytics attacks a
different issue or type of activity.
Basic rules on prior known schemes
have their place and cast a wide net for
detecting suspicious activity. An example
would be, “Show me all claims where
the patient is traveling more than 200
miles for routine care.” But rules are very
linear in the way they are written, and
investigators end up chasing many false
positives to weed out the cases that may
be explainable. Fraud rules are also very
easily gamed, once the fraudster figures
out how the rule is written.
Anomaly detection, which casts the net
a bit wider and looks for oddities you
didn’t know enough about to write a
specific rule around. However, such
oddities show up as outliers in behav-
iors that don’t look normal. Fraudsters
can also game anomaly detection,
for example, by working with multiple
providers to make their schemes harder
to detect through this method.
Predictive modeling applies statistical
methods like decision trees and neural networks to data from prior cases of known
fraud, and looks for statistical similarities.
A good example is a chiropractor who
treated everyone in the same family for the
same low back pain twice per week until
all family members’ health plan benefits
were exhausted. You can build a model
that can look for similar characteristics to
detect similar fraud schemes before they
are paid. Predictive models will vastly reduce the false positives and make analysts
laser-like in the claims they pursue for full
investigation.
1
2
Social network analysis (also called link
analysis) builds mathematical models that
show the connectedness of different entities
and score their statistical significance for
fraud, either by looking at their activities or
by looking at their personal relationships.
Physicians who always send patients to a
certain lab or two physicians who attended
the same schoolwould be linked entities,
for example. Once entities are linked, it
becomes interesting to see which grouped
entities violate a rule or model from one of
the other methods previously described.
This method allows analysts to spot
organized crime and collusive behavior.
Valuable data sources
often ignored
Medical claims data is the first
and strongest data source for fraud
investigations, but you can improve
fraud detection efforts by including
some of these additional data sources:
•M
ember eligibility data to compare
benefit coverage dates, falsification of
service dates, or falsification of supply
purchases. Past medical provider history,
including previous sanctions against
providers, their presence on state
and federal “watch” lists, and any
underlying motives for deception.
•A
ncillary claims like pharmacy billings,
lab data and hospital records with
revenue can detect whether all the
care provided naturally seems to fit
together as compared with accepted
medical practice.
•S
tructured and unstructured text data,
including nurses’ notes, claim processor
notes in claim records, electronic medical
records and call center logs, provide a
valuable and rich source of investigative
data that may be locked in a payer’s
underlying systems.
Proven value of
unstructured text data
To understand the value of text analytics
in detecting fraud, consider the text in
call center logs. Wouldn’t it be interesting
to know if the same chiropractor’s office
called the health plan five times in one
week to determine remaining benefit levels
on every family member covered under the
plan? By mining text data in the call center
logs, this could be flagged as an outlier.
And the nurses’ notes? Wouldn’t it be
meaningful if the notes documented a
physician saying this patient needed a
five-day hospital stay rather than two days
due to multiple comorbidities that are not
documented in the claim? The physician is
either omitting information in one source
or fabricating “facts” in another.
Finally, what might an audit of electronic
medical records reveal? Perhaps there
are far too many patients who all have
the same documented height, weight and
symptoms to be a normal occurrence. It
could alert the investigator that someone is
copying and pasting information, instead
of filling out medical records accurately.
also prepare a better bank of evidence
if the case ever does proceed to law
enforcement or prosecution.
It is becoming a recognized necessity
for health plans to find fraud before the
money goes out the door. The window
of opportunity is shorter than ever due to
timely payment guidelines and expectations
from regulators and the marketplace.
Investigators and analysts that use
anomaly detection, predictive modeling
and social network analysis – combined
with access to all sources of relevant
data–will have the best chance of
moving swiftly and accurately to detect
fraud, waste and abuse before the
loss occurs.
online
Read the SAS text mining blog:
blogs.sas.com/text-mining
The punch line
Using multiple analytical methods and all
the varied data sources available allows
organizations to conduct more efficient
investigations, know about fraud and
abuse cases sooner, and find out pertinent
information quickly. The investigator can
Julie Malida is the Principal for Health Care Fraud in the Enterprise Financial Crimes Global Practice at SAS. She has
devoted 29 years to the health care industry, focusing on managed care, fraud and cost containment in medical
claims. Malida is also a Fellow of the Society of Actuaries and a member of the American Academy of Actuaries.
1
3 Hyundai Marine &
Fire Insurance
prevents fraudulent
auto claims
Avoiding losses saves costs for insurer
and innocent policyholders
Korea’s automobile insurers are getting
slammed with fraudulent claims. The
country’s Financial Supervisory Service
reports a one-year increase of 30
percent. Unfortunately, most insurers do
not respond effectively – either because
they lack enough claims investigators
or because they base their predictive
models on existing industry statistics.
But not Hyundai Marine & Fire Insurance.
Using SAS for fraud detection and
prevention, Korea’s largest non-life
insurer has built a system that prevents
claims fraud and improves premium
payment processing, thereby protecting
its most profitable customers.
“Fraud costs not only the insurance
companies, but also the innocent
policyholders,” explains Chul-Woo
Lee, Manager of Hyundai M&F’s
Investigation Team. “For an insurance
company, enforcing the existing fraud
prevention system is no less important
than creating new profit.”
Hyundai M&F’s fraud detection system
combines business rules based on the
experience and knowledge of its
investigators with model rules generated
from data extracted from various IT
systems. The models are applied to
insurance claims, delivering results to
claims investigators in real time.
The system also monitors the performance
of these rules, enabling the company to
quickly modify existing rules or generate
new ones. The predictive models enhance
the process of fraud detection so Hyundai
M&F can detect fraudulent claims before
they ever get paid.
1
4
“It was inconvenient and ill-timed for us
to have to investigate every individual
accident history and assess the risk
information by going on the network
data every time,” Lee says. “Our fraud
detection system helped us to increase
the detection rate by technically securing
twice the amount of data than before.
And as a result we were able to secure
competitiveness in statistical service.”
Hyundai M&F’s system relies on a threepronged approach. A fraud pre-detection
system helps investigatorsjudge the
possibility of fraudulent activities using
statistics-based model and business rules,
allowing a loss assessment in advance
for the filed claim. A post-detection system
pinpoints potential fraud in areas where it
is hard to detect earlier, such as hospitals,
repair shops and individual clients. Lastly,
a risk mart generates data for both systems
in a separate data warehouse.
“For an insurance company, enforcing the
existing fraud prevention system is no
less important than creating new profit.”
Chul-Woo Lee,
Manager of the Investigation Team
The benefits of Hyundai M&F’s
system include:
• Timely risk assessment.
• Timely payout after selective
investigation.
• Better detection and accuracy.
• Easier gathering and maintenance
of fraud data.
Hyundai M&F also uses SAS to analyze
customer complaints. And the insurer is
moving toward a consolidated risk
management system.
“Fraud and risk management have
a lot in common,” Lee explains,
“so it’s important to have
complementary systems.”
online
SAS for fraud detection and prevention:
sas.com/industry/ins/fraud
Korea’s top insurance companies
predict fraudulent claims
using SAS
®
Korea’s insurance companies are struggling with preventing fraudulent claims and
enhancing customer satisfaction. The country’s Financial Supervisory Service reports a oneyear increase in fraud of 30 percent. Unfortunately, most insurers do not respond effectively,
either because they lack enough claims investigators or because they base their predictive
models on existing industry statistics.
SAS Korea is helping insurance companies address risk and fraud to prevent insurance
leakage, protect good subscribers and reducing the time for premium payments. Consider
Samsung Life Insurance. Using SAS, the company built a risk scoring system, an automatic
fraud detection system, a credit data management system, and a document forgery and
alteration detection system – first-of-their-kind industry implementations in Korea.
Following the success with Samsung Life Insurance, SAS moved on to help Hyundai Marine
& Fire Insurance, Korea Life Insurance, Kyobo Life Insurance, Dongbu Insurance and
Samsung Fire & Marine Insurance, all top 10 insurance companies in the country.
1
5
The future of fraud
investigations
Alison Bolen Intelligence Quarterly, Editor in Chief
In many ways, the early 1990s were the
heyday of Special Investigation Units (SIU)
for insurance companies, says Tim Wolfe,
Director of Special Investigations for CNA,
a commercial property and casualty
insurance provider. Around that time, states
started requiring companies to report
suspicious claims and insurers increased
staff to meet those regulations. “In those
days, SIU consisted almost entirely of
former law enforcers. We were hiring
people who were used to policing crime.”
field investigators asking the right
questions.” Technology does not do
investigations for you, but technology
will identify potential fraud activity that
may have been overlooked so you know
where to send those investigators and
waste less of their time.
Today, analytics technologies that identify
fraudulent activity are changing all of that.
Insurance companies are hiring data
modelers instead of security professionals,
and they’re changing the way SIU
departments are organized.
The staffing change is reflected at CNA
dramatically. In fact, Wolfe’s team has
outsourced a large portion of its field
investigative work and now employs a
handful of workers who manage and
monitor the investigative process, plus an
in-house team of investigators for major
investigations and organized crime.
The rest of his team is focused on intake,
training, regulatory compliance, analytics
and process improvement.
“That doesn’t mean you don’t need boots
on the street,” says Wolfe. “You still need
“The last data analyst we hired was from
the military,” says Wolfe. “Her experience
was in predicting IED explosions, but we’ve
found that it was a really good idea to hire
somebody outside of the industry. They’re
not just thinking insurance and they bring a
lot of fresh ideas.”
The change in skill sets is reflected at a
national auto insurance company in the
U.S., where a senior manager of the
Special Investigations Unit tells us they
have transformed the home office
environment over the last couple of years
to be an innovation environment. As a
result, his team is really starting to look
at how data can influence decisions.
Using data to identify fraud
Data has been used traditionally in
underwriting and claims side of most
insurance businesses, but it’s a recent
practice to use data to manage resources
and flag potential fraudulent claims for
“The last data analyst we hired was from
the military. Her experience was in
predicting IED explosions, but we’ve
found that it was a really good idea
to hire somebody outside of the industry.
They’re not just thinking insurance and
they bring a lot of fresh ideas.”
Tim Wolfe, Director of Special Investigations for CNA
investigation. There are clear, distinctive
data points that can be used for modeling. If businesses can identify these data
points and convert them into operational
decisions while collecting information from
customers, they can be more predictive.
In the past, most fraud referrals came from
a small percentage of adjusters who know
what to look for, take the time to make
reports, or have good instincts. Today’s
fraud systems, however, can identify red
flags automatically to help make adjusters
and SIU unit aware of potential fraud more
often and more quickly.
Another fortunate turn in identifying fraud
in insurance is that regulators are now
allowing insurance companies to share
info as it relates to fraud. For example, a
Medical Crimes Database is being built
with impetus from the National Insurance
Crime Bureau in the U.S. where all major
insurance companies can share medical
code records (in aggregate without
revealing individual patient info) to better
detect fraudsters who are making claims
across multiple insurance companies.
The value becomes obvious when you see a
single person claiming 60 hours of work in
a 24 hour day. If you’re only looking from
one company, you might see five hours of it
– but looking across every claim, the crime
becomes obvious.
Wolfe says technology is also important
because fraud is getting more sophisticated
than it used to be. “We’re seeing more
granulated schemes involving not just
doctors and lawyers but people associated
with workers compensation who bill for
interpretation services when they’re not
needed. Or transportation providers to
drive patients for care when no
transportation has took place.”
Ultimately, identifying fraud helps
everyone. It reduces business costs,
which can be passed on to the customer
and it helps fight larger societal criminal
elements. Plus, claims from legitimate
customers get processed more quickly
and fewer false positives keep honest
customers happier too.
online
Stop Fraudsters Before They Strike webinar:
sas.com/reg/web/corp/1478025
1
6
Preventing
fraud is in
everyone’s
best interest
While the true dollar impact of fraudulent
claims is almost impossible to measure, the
Insurance Information Institute estimates
that it represents approximately 10 percent
of all property/casualty insurance claims.
From 2005-2009 in the United States,
the dollar impact of that fraud totaled $30
billion. If you think the insurance company
just writes that off and it doesn’t affect you,
you are wrong. Fraudulent claim costs are
in many ways passed down to all policyholders in the form of increased premiums:
You’re paying for the bad behavior of
others. So, it’s in everyone’s best interests to
limit fraudulent activity: Insurers, regulators
and policyholders.
Learn from industry expert Jodi Pratt
how to create an anti-fraud culture:
sas.com/knowledge-exchange/risk/
fraud-financial-crimes/creating-ananti-fraud-culture/
Alison Bolen is the Intelligence Quarterly Editor and the Editor of the SAS blog
program. Since starting at SAS in 1999, Alison has edited print publications, websites,
e-newsletters, customer success stories and blogs.
1
7
How to prevent
fraud in the Indian
telecom industry
Understand the different types of fraud
and what you can do to combat them
Amitava Ghosh Senior VP, Analytics Platform Technology, Wireless Business, Reliance Communications
The Indian telecommunications industry is
the second-largest and fastest growing in
the world. According to a March 2012
Telecom Regulatory Authority of India (TRAI)
report, there were more than 919 million
wireless subscribers in the country with an
annual growth rate of 0.88 percent. During
that same time period, annual telecom
fraud worldwide was estimated at US$40.1
billion, according to the Communications
Fraud Control Association (CFCA) Fraud
survey for 2011.
Globally, telecom fraud decreased 33
percent from 2008 to 2011. However,
fraud in Indian telecom, as well as other
industries, is rising. According to a KPMG
Fraud Survey Report from 2010, 75
percent of respondents from all industry
sectors believe fraud incidence has
increased in India over the past two years,
54 percent feel that fraud is on the rise
in their own industry and 45 percent
feel that fraud has increased in their
own organization.
Sales and distribution fraud:
• Impersonation and
identity duplication.
To prevent fraud, 13 percent of survey
respondents said they are using IT controls
for fraud detection and 21 percent are using
data analytics to detect fraud.
Network and device fraud:
• PBX fraud.
• Subscription fraud.
• SIM box/interconnect fraud.
• SIM card cloning.
A growing marketplace presents many
challenges, including learning how to curb
the fraud and abuse that naturally multiply
as the market size increases. In this article, I
will classify fraud into three main areas and
describe different methods for combating
fraud for each type. The categories are:
Billing fraud:
• Unauthorized prepaid to
postpaid conversion.
• Suppression of billing
call detail records.
• Billing configuration.
Sales and distribution fraud
KPMG estimates that 70 percent of telecom fraud in India occurs because of flaws in the subscriber acquisition process, while 10
percent of revenue losses occur due to subscription fraud. Even though these types of fraud are considered entry-level, their sheer
prevalence adds up.
Impersonation fraud or identity duplication is a common entry-level fraud in India. Most customers in India subscribe for service
through a retailer. While retailers are required by law to verify the subscriber documentation (CAF), the retailer or a subscriber
is able to use the same document repeatedly, sometimes just changing part of the name or address. With each acquisition, the
retailer makes a significant commission.
SOLUTION: Use good de-duping software to match the CAF against the subscriber. Telecom companies can then check for
multiple connections and, if duplication is discovered, terminate the subscription and withdraw the commission from the retailer.
If a history or pattern of fraud is uncovered, offending retailers and subscribers can be blacklisted.
Another prevention method is to check for the reuse of International Mobile Equipment Identity (IMEI) numbers. Since most SIM
cards are activated by sending an SMS as part of the activation process, the IMEI number of the handset can be captured by a
fraudster. Single handsets used to activate multiple SIM cards within a certain time span can indicate fraudulent activity.
Subscription fraud, another entry-level fraud that is prevalent in India, happens when a retail agent or call center agent attaches a
value-added service (VAS) to an unsuspecting subscriber. For example, a ringtone can be added without the customer’s knowledge
or permission, resulting in a commission for the agent.
SOLUTION: Check the call detail records (CDR) to determine if a subscription request was made by the subscriber via SMS or by
contacting the call center. These kinds of audits should be performed regularly.
Network and device fraud
PBX fraud involves using PBX access codes to make outgoing calls to international destinations. PBX fraud occurs when
trespassers access business phone systems that use a PBX phone system and illegally use an access number to gain a dial tone.
They often place multiple long-distance calls through these lines for unscrupulous purposes, including the reselling of long-distance
for a profit. Service providers often cannot distinguish these calls from any other call originating from that business.
SOLUTION: Monitor calls to hot-listed international destinations that you know are popular with fraudsters. You should also
monitor the length of international calls.
SIM box and interconnect fraud is perpetrated to avoid call charges. A SIM box is a device that maps an international voice over
Internet protocol (VoIP) call to a SIM card of the mobile operator on the receiving end of the call. By doing this, the international
call appears to the destination operator as a local domestic call rather than an international call, thus cheating that operator out
of the international terminating charges, which can be significant. The same technique can be used to make a local out-of-network
call appear as if it were in network, resulting in significant revenue losses since out-of-network calls have termination charges.
SOLUTION: Telecom companies should be on the lookout for subscribers who have a very high number of distinct called parties
but too few distinct base transceiver stations. Also, look for a series of calls where roaming is very low or nonexistent and watch
for international calls with masked domestic numbers.
SIM card cloning fraud involves cloning the SIM cards of unsuspecting subscribers. By cloning the SIM card, the fraudsters have
free access to all of the services paid for by legitimate customers.
SOLUTION: Extensive call detail analysis must be done using data analytics techniques. For example, checking for call start
overlaps from the same multiple directory number is usually a good indicator once conference calls have been ruled out.
Billing fraud
Unauthorized prepaid to postpaid conversion fraud is generally committed by an employee of the telecom company who
removes the prepaid flag in Home Location Register (HLR), which then makes the billing postpaid.
SOLUTION: Monitor the system logs in the HLR for unauthorized access and changes. A service request that corresponds to the
conversion must be present. Audits should be performed very frequently to make sure the HLR configuration processes don’t have
weaknesses that can be exploited.
1
8
1
9
Suppression of billing call detail records (CDR) fraud involves suppressing the generation of CDRs in the network switch
and is usually carried out by a company employee.
SOLUTION: Control and monitor access to configuration switch processes to prohibit unauthorized activities. The revenue
assurance team must check for a decline in the number of CDRs generated from a switch based on past history.
Any significant drop in the rate should be examined.
Incorrect prepaid billing configuration fraud happens when an employee changes the configuration in the prepaid billing
systems called IN (Intelligent Network).
SOLUTION: A separate billing system is used, usually called Mediation, to take unrated CDRs directly from the
network switch and rate them by using the same billing configuration. This information should then be matched with
the data provided by IN. Any significant differences should be examined for fraud. Normally this is the function of a
revenue assurance team.
Analytics is the answer
Telecom fraud is big business around the world. With
rising competition and extremely low average revenue
per user (ARPU), the high cost of new acquisitions has
dramatically eroded the bottom line of telecom operators.
Detecting fraud and plugging revenue leaks have become
extremely important to reduce costs.
bottom line in the form of commissions and incentives.
All of the Asian and African prepaid telecom markets
are affected by each of the frauds mentioned above.
The most important weapon against prepaid fraud is data
analytics. At Reliance Communications, we use DataFlux
from SAS to de-dupe subscribers and to check for
impersonation fraud; we also use SAS Analytics to detect
subscription fraud. The revenue assurance team ensures
that billing and usage records reconcile. We are also
implementing a pilot project that will use subscriber history
to build profiles and check for anomalies in user calling
patterns that would indicate fraudulent behavior.
®
Fraud connected to prepaid accounts is much easier to
commit and harder to combat, since there is very little
information on the subscriber, unlike postpaid accounts,
where a credit check is usually done. Entry-level fraudulent
activities such as subscription and impersonation are very
serious since the cost is coming straight from the
®
Highlights on Telecom Subscription Data as of 31 March 2012. Telecom Regulatory Authority of India, 2012. trai.gov.in/WriteReadData/WhatsNew/Documents/PR-TSD-Mar03052012.pdf
Global Telecom Fraud Decreases by 33% from 2008, Returns to 2003 Levels. Communications Fraud Control Association, 2001.cfca.org/pdf/survey/Global%20Fraud_Loss_Survey2011.pdf
3
Ibid
4
India Fraud Survey Report 2010. KPMG, 2010. kpmg.com/IN/en/IssuesAndInsights/ArticlesPublications/Documents/KPMG_Fraud_Survey_2010.pdf
1 2
How fast is the Indian telecom market growing?
According to a March 2012 Telecom Regulatory Authority of India
(TRAI) report:
• Urban teledensity is 162.82. (Many urban people have
more than one phone.)
• There are more than 919 million wireless subscribers in India.
• Rural teledensity is 38.33.
• The growth rate in India’s telecom sector is 0.88 percent.
• India has the third-largest number of Internet users in the world,
with more than 121 million in 2011.
• Overall teledensity (number of wireless connections per 100
people) is 76.00.
5
Highlights on Telecom Subscription Data as of 31March 2012. trai.gov.in/WriteReadData/
WhatsNew/Documents/PR-TSD-Mar03052012.pdf
As a Senior Vice President at Reliance Communications, Amitava Ghosh implemented
a SAS analytic platform for the wireless business. He has many years of experience
constructing complex systems in all types of operating systems and a solid
background in building scalable, high-performance solutions.
Cutting tax
errors in half
2
0
Swedish Tax Agency finds erroneous tax returns
using predictive analysis
The Swedish Tax Agency has two
important objectives: to ensure that
taxes are correctly reported and that the
resulting tax debt is then paid. Using
new analytics solutions, the agency
can now identify risk profiles more
easily, thereby preventing and stopping
tax fraud and neglectful tax returns.
Statistician Joacim Danielsson. “This is
underpinned in part by good system
solutions where, for instance, employers
pay their employees’ preliminary tax and
submit a statement of income to the
Swedish Tax Agency. Even so, the aim
is still to ensure there are as few
discrepancies as possible.”
Like many other public sector organizations, the Swedish Tax Agency is facing
a future of diminishing resources.
“This increases the need for automated
solutions and requirements for better
resource management through wellfounded risk analysis,” says Andreas
Voxberg, Analyst/Section Coordinator at
the Swedish Tax Agency’s Analysis Unit.
In Sweden, the tax discrepancy, defined
as the tax that should be reported but isn’t,
is estimated to be roughly 10 percent of
the theoretically correct tax. One of the
Swedish Tax Agency’s goals is to half the
tax discrepancy, and one of the Analysis
Unit’s tasks is to identify anything that
could jeopardize meeting this goal.
e have high tax compliance in Sweden,
W
both in terms of reporting and paying it,”
explains Voxberg’s colleague, Analyst and
To reduce the percentage of tax errors,
the agency has endeavored to make life
easier for tax payers wherever possible,
gradually simplifying its tax returns.
It has also introduced various control
measures and conducted information
campaigns.
“ We have published information about
some of our controls online, so that
individuals and companies can run checks
themselves before submitting their returns
and avoid being flagged in our control
system. This saves time for them as well
as for the agency,” Voxberg explains. The
process of identifying tax returns with the
highest risk of errors takes place within the
framework of the Swedish Tax Agency’s
selection project, for which Voxberg and
Danielsson are project manager and
assistant project manager, respectively. All
tax returns are scrutinized. They undergo
an automated check, and some are also
selected for a more in-depth examination
according to certain criteria which may
vary from year to year.
2
1
“We need to
identify the
non-payers as
early as possible.
Using predictive
models, we can
identify risk profiles
and indicators so
that we can focus
where the risk of
non-payment
is greatest.”
Joacim Danielsson, Assistant
Project Manager, Analysis Unit,
Swedish Tax Agency
“All income and tax returns, from
individuals and companies alike, pass
through an auditing net, which is set
up by the Swedish Tax Agency’s
selection project. It is the project’s task
to identify risks at return level, based
on the operational plan and the various
activities therein. The risk analysis at a
more general level is dealt with by the
agency’s Analysis Unit, and is used as
a basis for decisions on what measures
should be implemented to deal with the
risk groups,” says Voxberg.
To produce as reliable risk profiles as
possible, the agency uses analysis solutions such as data mining and statistical
analysis. According to Voxberg, this is
not simply a means of “rating” tax
payers, rather it is a model for
pinpointing high-risk behavior.
Thanks to the agency’s new analysis
solutions, it has been able to broaden its
risk examination from the component to the
general level. Voxberg and Danielsson
regard the move from company level
to network level as the next fascinating
challenge. Putting the risk into a broader
context – a network – is a way of building
a more substantial picture.
The aim is to further complicate tax fraud.
When it comes to the risk of non-payment
of tax, this has become an increasingly
important area, particularly in the wake of
the credit crunch. Tax payment compliance
is at 99.7 percent, which means that only
0.3 percent is not paid.
It’s more about maintaining that level,”
says Voxberg, while noting that even 0.3
percent of Sweden’s total tax revenue
still equates to billions of kronor.
“We need to identify the non-payers
as early as possible. Using predictive
models, we can identify risk profiles and
indicators so that we can focus where
the risk of non-payment is greatest,”
says Danielsson.
Along with two master’s students from
Copenhagen Business School, the Swedish
Tax Agency is currently going through the
extensive research literature, so that, if
possible, it can improve its credit risk
models with further indicators of
inadequate solvency at both a company
and a macroeconomic level. Danielsson
can see many benefits in the improved risk
analysis, in addition to increased efficiency
for the Swedish Tax Agency:
It is good if we can find any debtors and
reach out to them early so we can help
out, and they can pay before the matter
is referred to the enforcement authorities.
Obviously it’s also a plus for society in
general if we have a stronger tax
foundation to fund welfare initiatives,
for instance.”
online
Stop fraudsters before they strike webcast:
sas.com/reg/web/corp/1478025
Reducing tax fraud
leads to better
customer service
in Belgium
2
2
Federal Public Service Finance uses analytics to
master huge fiscal database
Analyzing tax returns, verifying import
and export customs declarations, consolidating and checking the data from dozens of local tax collectors’ offices – those
are just a few of the jobs that Belgian
Federal Public Service (FPS) Finance has
entrusted to SAS software for reasons of
speed, efficiency and flawless execution.
FPS Finance is a government institution
that collects huge amounts of data from
a variety of sources. For a start, there’s
the Collection and Recovery Department
of FPS Finance that collects money from
taxpayers and pursues late payers. It
also recovers funds due from non-paying
individuals and companies, using all
possible legal means.
To give you an idea of the magnitude of
the operation: Every year, the department
handles information related to more than
3 million income tax returns, which results
in one of the largest databases in Belgium.
The department also manages the analysis
of real estate for estimating real values,
the analysis of all tax returns from legal
bodies as well from as private citizens
and the VAT, and the risk evaluation of
import and export customs declarations
for merchandise.
A perfect fit
“For the past few years, we have been
keen users of SAS Enterprise Guide
and SAS Enterprise Miner,™ mainly for
risk management in the various fiscal
domains,” says Dierk Op ’t Eynde, Data
Mining Coordinator at FPS Finance.
“The reason is very simple: They are
excellent solutions for data analysis and
data mining, two crucial missions for
our administration considering the size
of all the fiscal data to be analyzed, the
®
®
®
fast-changing Belgian legislation and the
European directives. And the nice thing
about it all is that the SAS tools fit perfectly
with the IT standards of FPS Finance.”
It used to be quite different altogether.
Efficiently managing and exploiting vast
databases of information is a challenge
for any system; processing times used to
be quite lengthy, which wasn’t exactly
conducive to higher efficiency or service
quality. That’s why in 2010 several
specialized services of FPS Finance
switched to a SAS environment for data
mining, risk analysis and performance
management.
The ICT department created a business
analytics environment in which a data
warehouse was directly linked to a risk
analysis environment, used by business
analysts from several departments for
2
3
a variety of operations, ranging from
data mining and risk analysis to ad hoc
inquiries. “Currently we have some 30
users supported by two administrators,”
Op ’t Eynde explains. “But in due time we
will increase this number by 30 percent
in view of the high expectations from our
policymakers regarding the handling of
fiscal fraud.”
Fighting fraud and more
Using SAS software to combat fiscal fraud
is a major undertaking, says Op ’t Eynde.
“It is our duty at FPS Finance to guarantee
a correct and fair tax collection, and that
means that every taxpayer should pay
what he or she legally owes, neither more
nor less. Fighting fiscal fraud is a major
focus here.”
Moving to SAS is
“like switching from
a handcrafted
data analysis
environment to an
industrial one.”
Dierk Op ‘t Eynde, Data Mining
Coordinator, FPS Finance
In addition to handling fraud detection
and tax collection, Op ’t Eynde says the
analytics are also used to improve customer
relationships with citizens and for creating
statistics, forecasts and simulations regarding all fiscal earnings. “Our use of SAS
solutions goes far beyond the fair collection
of taxes, a more efficient deployment of tax
inspectors or policy supporting information
for audits. Specific analyses will enable us
to recognize our stakeholders and enhance
our service to citizens and companies.
Our strategy is evolving toward a more
compliant and shared approach that
reflects the various risks and is primarily
aimed at preventing abuses.”
Apart from monitoring the tax collection,
risk analyses are carried out on a regular
basis to identify which individuals or
companies are at risk of not meeting their
fiscal liabilities, and which are likely to
become insolvent or bankrupt. Here, data
mining is put to good use to spot various
forms of fraud or to check on figures that
might help predict bankruptcies at an
earlier stage.
Previously it took the analysts up to half
a day to produce a scoring list, but they
can now obtain this list in a few minutes,
simply because the SAS tools are better
at handling huge volumes of data.
Op ’t Eynde called the move to the
SAS environment “like switching from a
hand-crafted data analysis environment
to an industrial one.”
User-friendly
One of the things that Op ’t Eynde
really appreciates is the fact that the
huge range of analytical options offered
by SAS is accessible via user-friendly
wizards. “On the one hand it means
there’s a fairly short learning curve for
new users. While on the other, more
experienced users enjoy the added
value of taking the next step to proper
development. Of course, it is all strictly
within the well-defined standards for
nomenclature, documentation and
version control.”
The same strict rules and regulations
apply to data quality. “Let’s not fool
ourselves,” Op ’t Eynde insists, “no single
administrative databank is perfect on
quality. That’s why, in order to get as
close to perfection as possible, we have
initiated two strategic programs. First,
there are major efforts underway to
modernize operational applications and
to make historical versions of the data
available for exploitation by building a
data warehouse. Second, we are
deliberately making great efforts to
ensure development, databases
and infrastructure are more uniform
and standardized.”
online
SAS Belgium and Luxembourg:
sas.com/belux
2
4
Dierk Op ’t Eynde, Data Mining Coordinator at FPS Finance
Where Does Your Institution
Rank in Fraud Prevention?
1
• Level 1 companies aren’t using any tools and can’t
measure the sources of fraud or the magnitude or
direction of payments risk.
However, the ad hoc tools used to acquire data
and manage fraud and risk may not be integrated
with one another to make cross-channel
communication and measurement effective.
• Level 2 companies have processes and tools, but
• Level 4 companies have a managed and meastill base investigations on intuition, tend to keep
sured approach. These institutions can benchmark
information siloed, and may not properly apply
themselves against industry performance, have
or understand the technology outside the immediprocesses in place to understand and root out new
ate project team. The debit card side of the house
types of fraud, and consider fraud potential prior to
often doesn’t know about the customer’s checking
entering new businesses.
account habits. And the rich demographic data the
marketing side uses to acquire and retain customers
• Level 5 companies are optimized. A portfolio
isn’t used to help understand the customer from a
approach is used to aggregate enterprise-level,
risk perspective. Measuring success isn’t common.
cross-channel payment risk. These companies are
• Level 3 companies have defined processes and use
completely up to date with regulatory compliance
tools to acquire data and assess fraud and risk.
and relevant regulatory guidance.
Learn more: sas.com/reg/wp/corp/31730
2
5
Using Analytics to
Predict Fraud
Irish tax officer describes how
Duncan Cleary Senior Statistician in Revenue, Irish Tax & Customs
Today, governments and their public sector
agencies everywhere are under pressure
to perform more efficiently and effectively;
essentially, doing better with less. Tax
and customs authorities are no exception,
many of which are dealing with decreased
resources and ever-increasing risks, often in
difficult economic circumstances.
So where does data mining fit into the
risk analysis toolkit of a tax authority?
Business rules aimed at detecting risk – and
the intelligence gathered from differing
channels – have their place and can
be effective. Add data mining to the mix
and you’ve got a powerful combination to
prevent and detect fraud and error.
Traditional methods for addressing risk
have served many authorities well, but
there is now a need to use more advanced
methods to combat fraud, error and waste.
To arm themselves for this battle, more and
more tax and customs authorities have
turned to data mining and analytics to
improve their business processes, resulting
in better compliance with new rules and
regulations and better customer service.
Data mining can be defined as the
application of the scientific method,
including statistical analyses, to large
amounts of data to uncover valuable
information from that data. It can often
detect patterns in data that cannot be
recognized manually, as well as make
predictive estimates of outcomes of
interest, such as the likelihood of a tax
return containing errors. Broadly speaking,
there are three types of data mining that
can be used to combat fraud:
1. Supervised techniques (also known
as predictive analytics), where a target
is predicted.
2. Semi-supervised techniques, where
some business knowledge can direct
the analyses.
3. Unsupervised techniques, such as
segmentation, which are exploratory.
Fighting fraud with
predictive analytics
By creating a predictive model, predictive
analytics uses a specific set of data that
contains known outcomes for a particular
target. This target could be likelihood to
yield if a case is audited, the likely amount
of yield, the likelihood of a business failing,
the likelihood of a claim for benefits or
refunds being fraudulent, and so on.
2
6
Models perform better where the target
has been clearly defined. Techniques for
creating predictive models are numerous,
but it is often hard to beat the wellestablished warhorses: logistic regressions,
decision trees and neural networks.
The real power of predictive models comes
from their ability to score new cases against
some target of interest, even if these cases
or events have never been previously evaluated. Cases can be ranked in descending
order of priority and worked according
to resources and the severity of the risk.
Feedback is critical to evaluating model
performance, and improving models is
an iterative and cyclical process. Feeding
information back into the model will help
reduce the number of false positives (false
alarms) over time, as well as reducing the
number of actual bad cases escaping
attention (false negatives).
Many tax agencies are now using these
predictive techniques, in conjunction with
their other tools such as business rules and
intelligence, to prevent and detect fraud
and error. Some have even deployed
these techniques in real time in their live
transactional systems, including the Irish
Tax and Customs authority.
Exploring fraud with
unsupervised techniques
Unsupervised techniques can be a
powerful means of understanding your
case base. Often there is so much data
available that it is difficult to understand
the underlying structure of the population
without using such methods as cluster
analysis and segmentation. Especially if
a target is not available, cluster analysis
can help identify groups in the population
that are alike within the group but different
from members of other groups.
Once segmented, cases can be assigned a
group membership. This label can be used
to determine treatment strategies, identify
service channel options and even monitor
the effectiveness of a tax authority’s efforts
to change taxpayer behavior over time.
Combining methods:
Semi-supervised techniques
Additional insight can be gained from
overlaying outputs from unsupervised
techniques with supervised techniques
and vice versa. Some segments might
emerge as inherently more risky, thus
semi-supervised techniques are often
useful where some business knowledge
can be used, even where only minimal
training data is available.
Outlier detection, where anomalies are
identified and investigated, also can be
an important weapon in the fight against
fraud. With any population, there will
always be anomalous cases, many of
which will be perfectly legitimate. However,
some cases may point to fraud, and these
can be identified and investigated.
The network view of the case base is
also becoming increasingly important in
detecting fraud and error, with group risk
and risk propagation through a network of
connected entities becoming increasingly
easier to detect using techniques such as
social network analysis. Unstructured data
– including text, voice, image and spatial
data – has just begun to be used for fraud
detection on a large scale, and its importance and usefulness will undoubtedly grow
in the future. Since tax authorities have the
unusual position of having population data,
not just sample data, there are few limits to
how data mining techniques can be used
to improve their performance.
Making analytics a core part
of your processes
So what is there to prevent the use of
data mining techniques in a tax authority?
There are many potential obstacles: lack
of good quality data, data that has not
been integrated or merged, lack of skilled
resources, lack of senior-level sponsorship,
IT challenges and cultural challenges.
Do not let these – or other issues – stop
you from using advanced analytics to
detect fraud, or stop you from using data
mining to improve an agency’s performance. Starting with a small achievable
project with a clearly defined goal can
often be the first step on the path to
success. The results do not need to be
spectacular, but if they show how data
mining can add value and potentially
reduce fraud and error, then the case
will be made and analytics can start to
become a core part of the agency’s
business processes.
Ultimately, it is the taxpayers and citizens
who will benefit the most if the public sector
adopts data mining as part of its day-to-day
business. So if analytics can help to reduce
fraud, error and waste, then the taxpayers
deserve nothing less.
online
SAS Ireland:
sas.com/ireland/
Duncan Cleary is a Senior Statistician in Revenue for Irish Tax & Customs. He specializes
in the use of research and analytics methodologies and their application in the Irish Tax
& Customs Authority, including the use of predictive analytics, customer segmentation,
risk analyses, large scale surveys, evidence-based decision support, social network
analysis and real-time risk.
4
7
9
11
13
Fraud detection is
more than cool
blogs.sas.com/content/
insurance
LA County closes in
on fraud rings
sas.com/solutions/fraud/
social-network
Preventing fraud
for the good of
society
sas.com/fraud-hybrid
sas.com/fraud-webcast
Is your health care
fraud detection
solution ignoring
valuable sources
of data?
blogs.sas.com/content/
insurance
Hyundai Marine &
Fire Insurance
prevents fraudulent
auto claims
sas.com/industry/ins/
fraud
15
17
20
22
25
The future of fraud
investigationS
sas.com/reg/web/
corp/1478025
How to prevent
fraud in the Indian
telecom industry
Cutting tax
errors in half
sas.com/reg/web/
corp/1478025
Reducing tax fraud
leads to better
customer service
in Belgium
sas.com/belux
Using Analytics
to Predict Fraud
sas.com/ireland
Journal of
Advanced Analytics
3Q 2O12
SAS Institute Inc. World Headquarters +1 919 677 8000
To contact your local SAS office, please visit: www.sas.com/offices
SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SAS Institute Inc. in the USA
and other countries. ® indicates USA registration. Other brand and product names are trademarks of their respective companies.
Copyright © 2012, SAS Institute Inc. All rights reserved. S96781.0812